Embed Size (px)
Citation preview
Understanding the representative gut microbiota dysbiosis inmetformin-treated Type 2 diabetes patients using genome-scalemetabolic modeling
Downloaded from: https://research.chalmers.se, 2020-10-11 17:48 UTC
Citation for the original published paper (version of record):Rosario, D., Benfeitas, R., Bidkhori, G. et al (2018)Understanding the representative gut microbiota dysbiosis in metformin-treated Type 2 diabetespatients using genome-scale metabolic modelingFrontiers in Physiology, 9http://dx.doi.org/10.3389/fphys.2018.00775
N.B. When citing this work, cite the original published paper.
research.chalmers.se offers the possibility of retrieving research publications produced at Chalmers University of Technology.It covers all kind of research output: articles, dissertations, conference papers, reports etc. since 2004.research.chalmers.se is administrated and maintained by Chalmers Library
(article starts on next page)
fphys-09-00775 June 21, 2018 Time: 15:56 # 1
ORIGINAL RESEARCHpublished: 25 June 2018
doi: 10.3389/fphys.2018.00775
Edited by:Kai Breuhahn,
Universität Heidelberg, Germany
Reviewed by:Clara G. De Los Reyes-Gavilan,
Consejo Superior de InvestigacionesCientíficas (CSIC), Spain
Marco Fondi,Università degli Studi di Firenze, Italy
*Correspondence:Saeed Shoaie
[email protected] Mardinoglu
†These authors have contributedequally to this work.
Specialty section:This article was submitted to
Systems Biology,a section of the journalFrontiers in Physiology
Received: 18 January 2018Accepted: 04 June 2018Published: 25 June 2018
Citation:Rosario D, Benfeitas R, Bidkhori G,
Zhang C, Uhlen M, Shoaie S andMardinoglu A (2018) Understanding
the Representative Gut MicrobiotaDysbiosis in Metformin-Treated Type
2 Diabetes Patients UsingGenome-Scale Metabolic Modeling.
Front. Physiol. 9:775.doi: 10.3389/fphys.2018.00775
Understanding the RepresentativeGut Microbiota Dysbiosis inMetformin-Treated Type 2 DiabetesPatients Using Genome-ScaleMetabolic ModelingDorines Rosario1†, Rui Benfeitas1†, Gholamreza Bidkhori1, Cheng Zhang1,Mathias Uhlen1, Saeed Shoaie2,3* and Adil Mardinoglu1,4*
1 Science for Life Laboratory, Royal Institute of Technology, Stockholm, Sweden, 2 Centre for Host-Microbiome Interactions,Dental Institute, King’s College London, London, United Kingdom, 3 Centre for Translational Microbiome Research,Department of Microbiology, Tumor and Cell Biology, Karolinska Institute, Stockholm, Sweden, 4 Department of Biology andBiological Engineering, Chalmers University of Technology, Gothenburg, Sweden
Dysbiosis in the gut microbiome composition may be promoted by therapeutic drugssuch as metformin, the world’s most prescribed antidiabetic drug. Under metformintreatment, disturbances of the intestinal microbes lead to increased abundance ofEscherichia spp., Akkermansia muciniphila, Subdoligranulum variabile and decreasedabundance of Intestinibacter bartlettii. This alteration may potentially lead to adverseeffects on the host metabolism, with the depletion of butyrate producer genus. However,an increased production of butyrate and propionate was verified in metformin-treatedType 2 diabetes (T2D) patients. The mechanisms underlying these nutritional alterationsand their relation with gut microbiota dysbiosis remain unclear. Here, we used Genome-scale Metabolic Models of the representative gut bacteria Escherichia spp., I. bartlettii,A. muciniphila, and S. variabile to elucidate their bacterial metabolism and its effecton intestinal nutrient pool, including macronutrients (e.g., amino acids and short chainfatty acids), minerals and chemical elements (e.g., iron and oxygen). We applied fluxbalance analysis (FBA) coupled with synthetic lethality analysis interactions to identifycombinations of reactions and extracellular nutrients whose absence prevents growth.Our analyses suggest that Escherichia sp. is the bacteria least vulnerable to nutrientavailability. We have also examined bacterial contribution to extracellular nutrientsincluding short chain fatty acids, amino acids, and gasses. For instance, Escherichiasp. and S. variabile may contribute to the production of important short chain fattyacids (e.g., acetate and butyrate, respectively) involved in the host physiology underaerobic and anaerobic conditions. We have also identified pathway susceptibility tonutrient availability and reaction changes among the four bacteria using both FBA andflux variability analysis. For instance, lipopolysaccharide synthesis, nucleotide sugarmetabolism, and amino acid metabolism are pathways susceptible to changes inEscherichia sp. and A. muciniphila. Our observations highlight important commensal
Frontiers in Physiology | www.frontiersin.org 1 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 2
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
and competing behavior, and their association with cellular metabolism for prevalentgut microbes. The results of our analysis have potential important implications fordevelopment of new therapeutic approaches in T2D patients through the developmentof prebiotics, probiotics, or postbiotics.
Keywords: gut microbiota, dysbiosis, host–microbiome interactions, genome-scale metabolic models, systemsbiology
INTRODUCTION
Dysbiosis in the gut bacterial community and concomitantmetabolic changes have an impact on human health (Qin et al.,2012; Tremaroli and Bäckhed, 2012; Karlsson et al., 2013;Forslund et al., 2015; Mardinoglu et al., 2016; Magnusdottiret al., 2017). Gut microbiome could affect host metabolism(Brillat-Savarin, 1826; Tremaroli and Bäckhed, 2012; Shoaieet al., 2013, 2015; Magnusdottir et al., 2017) through degradingnon-enzymatically digestible foods, and synthesis of aminoacids and short chain fatty acids (SCFAs). Dysbiosis may havedetrimental effects on host metabolism such as alterationsin abundance of nutrients crucial for homeostasis includingbutyrate (Forslund et al., 2015; Mardinoglu et al., 2016; Wu et al.,2017). Perturbations of intestinal microbiota are recognized as arisk factor for type 2 diabetes (T2D), a complex chronic disorderassociated with genetic and environmental risk factors such asage, diet, and lifestyle (Karlsson et al., 2013; Forslund et al.,2015; Shoaie et al., 2015; Mardinoglu et al., 2016; Magnusdottiret al., 2017). Recently, compositional shifts in representative gutmicrobes were identified in T2D patients undergoing metformintreatment, the most prescribed antidiabetic drug. These patientsdisplay increased abundance of Escherichia sp., Akkermansiamuciniphila (A. muciniphila), and Subdoligranulum variabile(S. variabile) (Forslund et al., 2015; Mardinoglu et al., 2016; Wuet al., 2017), and lower of Intestinibacter bartlettii (Forslund et al.,2015; Wu et al., 2017), as well as increased levels of the SCFAsbutyrate and propionate. Thus, despite potentially detrimentaleffects of gut microbiota dysbiosis, metformin-treated patientsdisplay beneficial alterations in gut SCFA abundances (Forslundet al., 2015; Mardinoglu et al., 2016). However, the relationshipbetween the metabolism of representative gut bacteria such asEscherichia sp., A. muciniphila, S. variabile and I. bartlettii, andcompounds in the intestinal lumen such as SCFAs or amino acidsis unclear.
Clarifying complex metabolic responses and relationshipsbetween gut microbes and host metabolism requires an analysisof large and highly intertwined reaction networks. GEnome-scale Metabolic models (GEMs) allow for the analysis of suchcomplex networks and have successfully been applied to clarifythe mechanisms underlying insulin resistance (Varemo et al.,2015; Zhang and Hua, 2016; Mardinoglu et al., 2018; Turanliet al., 2018) and to identify important nutritional interactionsbetween gut microbes and the host (Shoaie et al., 2013; Jiand Nielsen, 2015; Mardinoglu et al., 2015; Zhang and Hua,2016). Synthetic lethality analysis (Pratapa et al., 2015) isan approach commonly used in constraint-based modeling toclarify biological phenomena (Mardinoglu and Nielsen, 2012;
Mardinoglu et al., 2016; Magnusdottir et al., 2017). It is usedto identify vital interconnected metabolic processes underlyinga phenotype of interest (Qin et al., 2012; Shoaie et al., 2013;Magnusdottir et al., 2017) and has been extensively applied inhealth and disease (O’Neil et al., 2017). While synthetic lethalityanalysis traditionally seeks to identify genes that are individuallyessential, this approach may assist in identifying whether thesimultaneous knock-out of two genes of interest leads to celllethality, but their individual knock-out maintains cell viability,i.e., synthetic lethality interactions (Kaelin, 2005).
Through reconstruction and analysis of GEMs, we sought tounderstand the contribution of the four bacteria in the physiologyof T2D patients undergoing metformin treatment. We usedAGORA GEM reconstructions of Escherichia sp., A. muciniphila,S. variabile, and I. bartlettii to analyze relationships between thebacterial metabolism and the extracellular environment, as wellas predicting the survivability of the bacteria against nutritionalalterations (Magnusdottir et al., 2017). Here, we employed theconcept of synthetic lethality analysis to identify sets of individualand pairs of reactions that, when not present, abolished growth.Additionally, we implemented nutritional interactions analysisto understanding how the presence or absence of gut nutrientsinfluences bacterial growth by focusing on nutrient transportreactions (i.e., exchange reactions). Moreover, we assessed theinfluence of available nutrients and synthetic lethal reactions oncellular metabolic pathways to clarify which metabolic pathwayswere mostly dependent on nutritional alterations and undersurvivability threat, respectively. Lastly, interactions betweenthe gut microbiota and the environment (host intestine) wereevaluated through a novel approach based on the production andconsumption of substrates of interest under maximal growth andminimal media conditions of each organism. Our observationshighlight important association between cellular metabolismof these four prevalent gut microbes and point importantimplications for development of new therapeutic approaches inT2D patients.
MATERIALS AND METHODS
Genome Scale Metabolic ModelRetrieval, Curation, and ModelingAGORA (Assembly of Gut Organisms through Reconstructionand Analysis) model reconstructions (Magnusdottir et al., 2017)were downloaded in SBML format from Virtual MetabolicHuman (VMH) database1 for Escherichia sp. 4_1_40B, and
1https://vmh.uni.lu/#microbes/search
Frontiers in Physiology | www.frontiersin.org 2 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 3
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
I. bartlettii (Clostridium bartlettii DSM 16795) on the 27thof January 2017, for A. muciniphila ATCC BAA-835 andS. variabile DSM 15176 on the 2nd April 2018. Details regardingmicroorganism AGORA reconstructions are accessible inSupplementary Table S1. The models were manually curatedto ensure biological functionality. The computations wereperformed on resources provided by the Swedish NationalInfrastructure for Computing (SNIC) through UppsalaMultidisciplinary Center for Advanced Computational Science(UPPMAX).
The RAVEN (Reconstruction, Analysis and Visualization ofMetabolic Networks) Toolbox (Agren et al., 2013) was usedto define and set parameters for simulations and performanalyses of the originated predictions. Unless otherwise stated, allflux balance analyses (FBAs) considered biomass production asobjective function. For flux variability analysis (FVA), minimumand maximum flux ranges were calculated for each reactionfor the optimized value of the objective function throughthe COBRA (Constraint-Based Reconstruction and Analysis)Toolbox (Schellenberger et al., 2012).
Synthetic Lethality AnalysisLethality analysis was performed by adapting the Fast-SLalgorithm (Pratapa et al., 2015) from the COBRA Toolbox(Schellenberger et al., 2012) to RAVEN Toolbox (Agrenet al., 2013). Fast-SL-derived single and double lethal reactionspredictions (Supplementary Table S2) were further validatedby constraining methods, setting lower and upper bounds tozero, with biomass maximization defined as objective function.Single lethal reactions were determined and treated as essentialreactions for cell growth. Double lethal reactions were consideredas those pairs of reactions that induce no growth whenblocked simultaneously but not individually. Exchange reactionswere determined using default RAVEN functions, and onlythose involving nutrient exchange with the extracellular spaceare reported (i.e., outside reactions, and not inside reactionswhich include DNA replication, RNA transcription, proteinbiosynthesis and biomass, and are treated as intracellularreactions). This permits the identification of essential exchangereactions, which are the nutrients required to be uptaken fromthe environment by the organism in order to guarantee cellsurvival.
Metabolic Pathway Sensitivity toEssential Reactions and NutrientChangesThe built-in subsystems of the model were used for defining thepathways (Supplementary Table S5) and unclassified pathwayswere ignored. We applied modeling-constraints (lower andupper bounds set to zero and objective function definedas biomass maximization) going through each of the singleand double lethal reactions (essential reactions) and non-essential exchange reactions. Pathway sensitivity to changeswas determined based on the proportion of reactions thatpresented absolute flux changes above 0.01 mmol/gDW/hrelative to the respective flux in the reference model where no
constraints were set on lower/upper bounds. This value wasconservatively considered based on the observation that FBA-based approaches often use 0.001 mmol/gDW/h as thresholdfor identifying reactions that have fluxes (Hyotylainen et al.,2016).
Additionally, these results were compared with those fromFVA in response to the inhibition of single and paired syntheticlethal (essential) and non-essential exchange reactions, andcompared to a reference output without applied constraints onlethal neither exchange reactions. Only solutions on flux variationthat achieve ≥90% of the reference solution were considered.Using the minimum and maximum fluxes determined for eachreaction, we computed the mean and ranges for all reactions ineach subsystem.
Extracellular Nutrient Uptake andAlternative Aerobic and AnaerobicEscherichia sp. GrowthWe have employed a novel approach which allowed us toidentify which are the minimal sufficient nutrients that whencombined are capable of providing cellular growth when uptakenby the organism. In order to identify which nutrients are onthe first line promoting cellular growth under environmentallimited conditions, the target reactions of this approach werenon-essential exchange reactions. No constraint was appliedfor single essential exchange reactions to ensure that growthinhibition was not due to the block of required essential nutrients.Cellular intake through non-essential exchange reactions wasblocked with lower bounds set to zero. Based on FBA methods,non-essential exchange reactions were blocked one-by-one, two-by-two, and three-by-three. Future work should test how thisapproach compares with existing methods for determiningminimum growth conditions (e.g., Imielinski et al., 2006; Ekeret al., 2013). This was performed for all organisms underanaerobiosis, and also for Escherichia sp. under aerobiosis.A biomass flux threshold of 10−5 was defined as minimum toconsider cell growth.
Maximal Growth-Coupled ExtracellularNutrient Production and ConsumptionWe developed a novel approach to assess the contribution ofeach bacteria for nutrient production and consumption underthe maximum growth rate permitted under minimum mediaconditions. Specifically, we determined the maximal rate ofsecretion or intake of each metabolite when the organism is atits highest growth yield by individually setting each metaboliteof interest as objective function at a time, therefore maximizingits production or consumption. Maximum organism growthwas determined based on FBA under minimal media foreach organism (Supplementary Table S7). Thus, the predictedmaximal growth (0.6387, 0.2268, 0.2599, 0.2460 mmol/gDW/h,respectively for Escherichia sp., I. bartlettii, A. muciniphila, andS. variabile) was used as lower bound constraint for biomasstogether with minimal media conditions and under anaerobicconditions (with oxygen exchange constrained to zero in bothmodels).
Frontiers in Physiology | www.frontiersin.org 3 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 4
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
RESULTS
In Silico Identification of DifferentGrowth Requirements in RepresentativeGut BacteriaTo assess growth requirements of Escherichia sp., I. bartlettii, A.muciniphila and S. variabile, we retrieved AGORA (Magnusdottiret al., 2017) models for these organisms (SupplementaryTable S1). These models comprise the entire known metabolicreaction networks of these organisms, and contain 1757, 1095,1125, and 1057 reactions, and 1267, 730, 592, and 1313 genes,respectively. Using the FAST-SL algorithm (Pratapa et al., 2015)based on FBAs) with biomass as objective function, we performedsynthetic lethality interaction analysis (Figure 1) on these fourorganisms. Through this approach, we revealed the influenceof an inhibited (i.e., without flux) reaction on the metabolicnetwork. This allowed for the identification of single essentialreactions (Figure 1A), and those combinations of reactionpairs that become lethal when blocked simultaneously but notindividually (Figure 1B). In total, this represents between 559,153to 1,544,403 different conditions (including single and doublereaction combinations) tested.
Additionally, this approach allowed for understanding theconsequences of unavailability of environmental compounds(e.g., amino acids or oxygen) on cell growth by inhibitingtransport reactions with the extracellular environment (i.e.,exchange reactions). Escherichia sp., I. bartlettii, A. muciniphila,and S. variabile respectively displayed 211, 153, 142, and 130exchange reactions. I. bartlettii was the bacteria with higherproportion of exchange reactions, while S. variabile was theorganism with higher proportion of single lethal reactions.Escherichia sp. was the bacteria with lower proportion ofessential exchange reactions (Figure 1A and SupplementaryTable S2). These four organisms commonly shared 46 singlelethal reactions. A. muciniphila presented 182 organismspecific single lethal reactions and 45 additional single lethalreactions shared with Escherichia sp. Among all exchangereactions, 10 single-lethal were shared by these four organisms:environmental exchange of calcium, chloride, carbon dioxide,copper, potassium, magnesium, manganese, sulfate, zinc,and ferrous (Fe2+) iron (Figure 1C). Escherichia sp. did notpresent organism-specific essential exchange reactions, whereasI. bartlettii, S. variabile, and A. muciniphila respectively had 1,2, and 6 single-lethal exchange reactions found only in theseorganisms. Both A. muciniphila and S. variabile presented sharedsingle-lethal exchange reactions with I. bartlettii, where exchangeof ferric iron (Fe3+) was essential in the three organisms.Exchange of vitamin B5 and tryptophan were essential exchangereactions found in I. bartlettii and S. variabile, whereas exchangeof hydrogen phosphate is commonly essential in I. bartlettii andA. muciniphila.
When considering all possible pairs of combinations,A. muciniphila presented the highest proportion of double lethalreaction pairs and pairs that include at least 1 exchange reaction.I. bartlettii was the organism with higher number of organism-specific lethal reaction pairs with ≥1 exchange reaction, followed
by A. muciniphila (Figure 1B). There were no lethal reactionpairs (≥1 exchange reactions) exclusively shared by the twoorganisms. However, ornithine exchange comprised 7 and 4organism-specific double lethal reactions in I. bartlettii andA. muciniphila, respectively. Among those combinations foundtogether with ornithine exchange, the urea cycle was the onlycommon pathway between the two bacteria, where arginine andproline metabolism are specific for A. muciniphila, and alanineand aspartate metabolism, pyrimidine synthesis, citric acid cycleare specific for I. bartlettii.
Intracellular reactions involving NADP+/NADPH becamelethal when combined with riboflavin or diaminoheptanedioateexchange in I. bartlettii. In turn, intracellular reactions involvingNADP+/NADPH together with environmental exchange ofvitamin B5, the fatty acid laurate or thymidine were syntheticlethal reaction pairs in Escherichia sp., but not in I. bartlettii.Escherichia sp. displayed the lowest proportion of double lethalreactions, as well as the lowest proportion of double lethalreactions with ≥1 exchange reactions, and the lowest numberof organism-specific lethal reaction pairs with ≥1 exchangereactions. Among these pairs with ≥1 exchange reactionwhich involved fatty acids, Escherichia sp. and A. muciniphilarespectively displayed 20 of 21 shared reaction pairs involvinglaurate exchange and an intracellular reaction associated withfatty acid synthesis or oxidation. Simultaneous inhibition ofacetate exchange and acetate kinase or phosphotransacetylasereactions were lethal in A. muciniphila.
Several double lethal pairs involving nicotinate exchange werepresent in S. variabile (three pairs) and I. bartlettii (four pairs).Lethal pairs involving nicotinamide mononucleotide (NMN)exchange were also found for S. variabile (three pairs) andEscherichia sp. (two pairs), where one pair involves nicotinate-nucleotide adenylyltransferase in both organisms. Reactionsinvolved in hypoxanthine exchange and purine synthesis werefound in 10 double lethal pairs exclusive of S. variabile.
The inhibition of L-lysine exchange simultaneously withdiaminopimelate decarboxylase reaction was the only lethalpair found in the four organisms. Reaction pairs includingexchange of arginine, alanine, asparagine, aspartate, isoleucine,lysine, tyrosine, valine, thymidine, or thiamine (vitamin B1)were synthetic lethal in one or more organisms. S. variabile, A.muciniphila, Escherichia sp., and I. bartlettii respectively had 4,2, 1 and 1 double lethal pairs involving 2 exchange reactions.Simultaneous inhibition of NMN exchange and nicotinate, or L-tyrosine coupled with glycyl-L-tyrosine, or phosphate paired withglycerol 3-phosphate and uracil paired with succinate becamelethal in S. variabile. In A. muciniphila inhibiting the exchange ofL-asparagine together with glycyl-L-asparagine or thiamin led tolethality. Notably, simultaneous blocking of exchange of oxygenwith ferric iron (Fe3+) or nicotinate, respectively preventedgrowth in Escherichia sp. and I. bartlettii. While I. bartlettii isan obligate anaerobe (Supplementary Table S1), the observationthat O2 exchange is present in this model could indicate thatthe model failed to describe its aerotolerance. However, the twofollowing points indicate that the model predictions are robustto O2 availability. First, the Spearman correlation between modelfluxes in presence vs. absence of O2 was very high (Spearman’s
Frontiers in Physiology | www.frontiersin.org 4 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 5
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
FIGURE 1 | Specific growth requirements suggest lower vulnerability of Escherichia sp. to environmental nutritional deficiencies. (A) Proportions of single lethalreaction sets for the entire metabolic network and for exchange reactions in the four organisms, and number of exclusive and shared single lethal reactions(Supplementary Table S2) (B) Number of double lethal reaction pairs. (C) Essential metabolites consumed by the four organisms. Multiple colored metabolites areconsumed by several bacteria according to the legend.
ρ > 0.82, P < 10−70 considering all 304 non-null fluxes of bothmodels). Second, the reactions catalyzed by antioxidants againstreactive oxygen species (hydrogen peroxide reductase) showedactivity in a model encompassing oxygen exchange, but notin its absence (Supplementary Table S3). We finally, removedoxygen exchange from the I. bartlettii model and repeated thelethality analysis for the entire reaction network. The comparisonof synthetic lethality analysis under aerobic versus anaerobicconditions changes the number of single lethal reactions from80 to 85, and from 124 to 171 lethal pairs (SupplementaryTable S4), respectively. However, only one additional single lethalexchange reaction (methionine exchange) was identified in theI. bartlettii model. These observations reinforce the confidencein the predictions of the model in terms of environmentaldependency or synthetic lethal reactions.
Identification of Sensitive Pathways toInhibition of Lethal and Non-essentialExchange ReactionsWe investigated which pathways were mostly altered by singleand double synthetic lethal reactions (i.e., essential reactions)
and non-essential exchange reactions. Escherichia sp. displays 73metabolic pathways, I. bartlettii displays 66, A. muciniphila has68 and S. variabile displays 63, of which 54 are commonly presentin the four organisms (Supplementary Table S5). Consideringthe entire metabolic network and sets of single, paired essentialreactions, and non-essential exchange reactions individually andcoupled in pairs, we computed the proportion of reactions thatare altered in each pathway in comparison with each bacteria’sreference model, i.e., the model with no reaction blocking.To do so, we used FBA to identify flux distribution betweenpathways while maximizing for bacterial growth, i.e., “pathwaysensitivity” to reaction blocking. Additionally and to complementthis methodology, we employed FVA (Supplementary Table S6).We observed (Figures 2, 3) that several pathways show significantalterations in >50% of their reactions. For instance, cholesterol(and squalene) synthesis but not other reactions involved incholesterol metabolism, were highly perturbed by essentialreactions and partly by environmental exchange reactions inEscherichia sp. In turn, cholesterol metabolism was highlyperturbed by essential and environmental exchange reactionsunder the same constraints in A. muciniphila but not inI. bartlettii and S. variabile.
Frontiers in Physiology | www.frontiersin.org 5 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 6
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
FIGURE 2 | Pathways of Escherichia sp. and Intestinibacter bartlettii show distinct vulnerability to environmental nutritional changes. Synthetic lethality analysis wasperformed in Escherichia sp. (A) and I. bartlettii (B) for blocking of single reactions or pairs of reactions belonging to the entire metabolic network (“All essentialreactions”), for exchange reactions, or for non-essential exchange reactions and then we determined the fraction of pathway reactions altered (>1% change withrespect to the reference model). For reaction pairs, we also considered pairs comprising 1 exchange reaction and 1 intracellular reaction. Columns have differentnumber of blocked reactions, and only one pair of essential exchange reactions was found in each organism (see text). Columns leading to no pathway changes arenot shown; for non-essential exchange reactions, only those in the top 30% inducing most pathway changes are shown. Amino acids are abbreviated by theircommon three-letter names. Total number of reactions in each pathway are presented in brackets. Due to the large number of possible combinations, only thosepairs of non-essential exchange reactions that resulted in high pathway changes are shown (sum over all pathways >4).
Frontiers in Physiology | www.frontiersin.org 6 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 7
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
FIGURE 3 | Pathway susceptibility to environmental changes in Akkermansia muciniphila and Subdoligranulum variabile. Synthetic lethality analysis was performed inA. muciniphila (A) and S. variabile (B) similarly to Figure 2. Amino acids are abbreviated by their common three-letter names. Total number of reactions in eachpathway are presented in brackets. Due to the large number of possible combinations, only those pairs of non-essential exchange reactions that resulted in highpathway changes are shown (sum over all pathways >4).
Frontiers in Physiology | www.frontiersin.org 7 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 8
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
Akkermansia muciniphila showed the most significant cellularpathway alterations in response to essential reactions includingN-glycan synthesis, exclusive to this bacteria (Figure 3A).Lipopolysaccharide (LPS) biosynthesis and nucleotide sugarmetabolism were metabolic pathways highly perturbed inA. muciniphila and Escherichia sp., but not in S. variabile. I.bartlettii showed (Figure 2B) substantial (>50%) alterationsin metabolism of propionate, phenylalanine, alanine but nochange in chloroalkane and chloroalkene degradation, a species-exclusive metabolic pathway. In turn, metabolism of butyrate andvitamin B2 showed substantial (>50%) alterations in metabolismin Escherichia sp. (Figure 2A). Oxidative phosphorylation, ametabolic pathway found in the four organisms, was highlyperturbed in S. variabile when any of its lethal or non-essentialexchange reactions were inhibited. The same was observed inI. bartlettii, however mainly when double lethal reactions wereblocked simultaneously. The metabolism of sulfur and energywere equally highly perturbed in S. variabile in response toinhibition of any of its essential or non-essential exchangereactions, while sulfur metabolism was poorly affected in otherspecies and energy metabolism was only considerably perturbedin A. muciniphila.
While some of the pronounced changes exhibited by somepathways reflect their small size (e.g., oxidative phosphorylationwith ≤3 reactions), other pathways showed substantial changesthough they comprise more reactions. This is the case of LPSbiosynthesis (27 reactions in Escherichia sp. and 30 reactions inA. muciniphila), butyrate metabolism (9 reactions in I. bartlettiiand Escherichia sp.), or phenylalanine metabolism (25 and 10reactions in Escherichia sp. and S. variabile). Importantly, thesetrends were also observed when blocking single or pairs ofessential exchange reactions, and for many of the non-essentialexchange reactions, indicating the strong effect of nutritionalavailability in these pathways. Metabolic pathways were moresensitive to inhibition of essential (lethal) reactions that areintracellular and environmentally exchanged, comparatively tonon-essential exchange reactions in the four organisms. FVAshowed qualitatively similar results, though it indicates that morepathways were sensitive to perturbations than FBA.
Tyrosine, Phenylalanine, and Vitamin B6Permit Escherichia sp. Growth UnderAerobic but Not Anaerobic ConditionsBacteria present different growth requirements, and thus maypresent selective advantages and disadvantages. Among the fourbacteria tested here, all are strict anaerobes with exception toEscherichia sp., a facultative aerobe. In Escherichia sp., blockingof oxygen and iron exchange together induces lethality (but notindividually, since production of ferric iron depends on oxygenthrough the reaction 4H++ O2 + 4Fe2+
→ 2H2O + 4Fe3+).We questioned if pathway utilization may differ not only inresponse to nutrients but also in response to oxygen availability(Figure 4A, top). Such differential nutritional responses maypresent an added selective advantage over anaerobic bacteria.
We investigated pathway response to oxygen availabilityin Escherichia sp., and determined the minimum growth
requirements for the four organisms. We developed an approachcomplementary to those used above for assessing pathwaysensitivity (Figure 4A, bottom). Briefly, from the entire metabolicreaction network, we selected those involving exchange reactionsand blocked all non-essential single exchange reactions identifiedabove, whereas the single-lethal exchange reactions identifiedabove are unblocked. All non-lethal exchange reactions are firstlyblocked, and then unblocked one by one, two by two, etc. Thesynthetic lethality approach employed above optimized for cellgrowth, and thus allowed for identification of those exchangereactions that most penalize cell growth and whose blockingprevents cell growth using otherwise unconstrained models.In turn, the approach used here optimizes flux distributionin a tightly constrained model and permits identifying thosecombinations of exchange reactions that, when simultaneouslyunblocked, promote cell growth. This additionally permitsidentifying those pathways showing the greatest changes whileconferring the greatest increments to cell growth by comparisonwith the reference fully unconstrained model.
None of the four gut bacteria under study displayed cellulargrowth when unblocking any single exchange reaction. Onlypairs comprising either oxygen or iron exchange resulted ingrowth for Escherichia sp. when combinations of two-by-twonon-essential reactions were allowed. In combinations of threeexchange reactions, oxygen and iron exchange is always presentas one of the necessary reactions for growth (results not shown).Unblocking combinations of two non-essential reactions inA. muciniphila and S. variabile provided significant cellulargrowth. Notably, employing this approach yielded no growth inI. bartlettii using combinations of 1, 2, and even 3 unblocked non-essential exchange reactions (results not shown), suggesting thatmore nutrients must be available in order to permit growth. TheEscherichia sp. model shows growth with as few as 12 exchangereactions (of which 10 are single-essential), whereas the model forI. bartlettii does not grow with 18 exchange reactions (including15 single-essential). Additionally, A. muciniphila shows growthwith only 23 exchange reactions (which includes 21 single-essential), while S. variabile shows growth with only 14 exchangereactions (of which 12 single-essential).
Escherichia sp. displays substantial pathway changes(Figure 4B) in LPS, squalene and cholesterol biosynthesis,nucleotide sugar metabolism (>90% pathway reactionswith >1% fluxes under all assessed conditions), purine andbutyrate metabolism (>74% reactions altered), as well asmetabolism of histidine, tryptophan, valine, leucine, isoleucine,aspartate, alanine, and lysine (>70%). Glutathione and nitrogenmetabolism tend to be mostly unchanged (<2%). Additionally,we observed that Escherichia sp. responds differently dependingon oxygen availability. As expected from aerobic growth,ROS detoxification is significantly active when O2 exchangeis unconstrained versus no changes when Fe3+ is unblockedbut O2 exchange is blocked (respectively, >33 vs. 0%, compareFigure 4B, left with right). Slight increased fluxes are alsoidentified under aerobic conditions in energy metabolism(mean pathway reaction changes >21% aerobic vs. 16%anaerobic), pentose phosphate pathway (57 vs. 50%), starchand sucrose metabolism (14 vs. 3%), and metabolism of
Frontiers in Physiology | www.frontiersin.org 8 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 9
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
FIGURE 4 | Effect of availability of environmental nutrients on pathway responses and growth rates in Escherichia sp. under aerobic and anaerobic growth. (A) Dueto their aerotolerance, the facultative Escherichia sp. may respond differently to environmental nutrients under aerobic versus anaerobic growth, which may provide aselective advantage with respect to the obligate anaerobe I. bartlettii. We employed a novel in silico approach where all single essential reactions are kept unblocked,and the non-essential exchange reactions (in Figure 2) are unblocked one-by-one and two-by-two. This approach may thus assist in identifying those combinationsthat confer the highest increments on cell growth, as well as determining which pathways most support this response. (B) Escherichia sp. pathway reactions that arealtered (% from total) as response to availability of specific nutrients, together with oxygen or iron exchange (only pairs that either included oxygen or iron exchangeresulted in growth). No growth is observed when unblocking single reactions for Escherichia sp., or in I. bartlettii for single, pairs, or triplets of reactions (results notshow). Reactions were considered altered when their flux were altered >1% against the reference model. (C) Growth rates for Escherichia sp. achieved byunblocking exchange reactions together with O2 exchange (i.e., aerobic conditions) or iron exchange (i.e., anaerobic conditions).
Frontiers in Physiology | www.frontiersin.org 9 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 10
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
vitamins B6 (64 vs. 57%) and B2 (71 vs. 64%). Oxidativephosphorylation shows substantial increases (>66%) in someentries under aerobic conditions, but not under anaerobicconditions, when exchange of some compounds and amino acidsis unblocked (e.g., NMN, alanine, glutamine, glycine, proline,serine, threonine, and tryptophan). In turn, sulfur metabolism(18 vs. 25%) and glyoxylate/dicarboxylate metabolism (29vs. 35%) are slightly altered under anaerobic conditions.Practically all nutrients confer highest growth rates underaerobic than anaerobic conditions, with exception to nitrateexchange that elicits similar growth rates under aerobic andanaerobic growth. NMN, glutamate, aspartate, nitrate, andnitrite exchange confer the most substantial increases togrowth under aerobic and anaerobic conditions (Figure 4C).Interestingly, tyrosine, phenylalanine, tyramine, and vitamin B6uptake allow for cell growth under aerobic but not anaerobicconditions.
Commensal and Competing MetabolicBehavior of Gut Bacteria in theUtilization of Amino Acids and ShortChain Fatty AcidsWe also determined how amino acids, short chain fattyacids, and other nutrients important for host and bacterialmetabolism (Shoaie et al., 2013; Forslund et al., 2015; Mardinogluet al., 2016) were produced or used by the four bacteria.To this extent, we aimed to determine for each metaboliteits maximum growth-coupled uptake/secretion fluxes undermaximal growth and minimal media conditions in anaerobiosis(Supplementary Table S7, see section “Materials and Methods”),since the human intestinal environment is predominantlyanaerobic (Tremaroli and Bäckhed, 2012; Donaldson et al.,2015). We observed that the four organisms may contributefor the production of extracellular acetate, whereas all butS. variabile produced propionate. Predictions have shownbutyrate production by S. variabile. In turn, I. bartlettii producedisobutyrate (Figure 5A and Supplementary Table S8), whileboth Escherichia sp. and I. bartlettii revealed to compete forribose, deoxyribose and cysteinylglycine, as well as for aspartateand phosphate, which were both products of S. variabile(Figure 5B).
Potential commensal behavior may occur, since some ofthese compounds may be produced by A. muciniphila andS. variabile (e.g., threonine and glycine), while consumed byEscherichia sp. and I. bartlettii. Phenylalanine produced bythe three other bacteria may be consumed by I. bartlettii,which in turn is predicted to secrete phenylacetate. Prolineand glutamine were produced by A. muciniphila, I. bartlettiiand S. variabile and consumed by Escherichia sp. Finally,Escherichia sp. was involved in the production of the gasseshydrogen and, together with A. muciniphila, both may producehydrogen sulfide; whereas I. bartlettii produced methanethiol(Figure 5C). Because Escherichia sp. is a facultative aerobewe repeated these analyses under aerobic conditions, andobserved some differences in comparison with the resultsunder anaerobiosis, specifically in the secretion of amino
acids (e.g., proline, glutamate, and threonine) and nucleobases(Figure 5B).
Altogether, our results demonstrated that the four bacteriadisplayed substantial differences in substrate requirements forgrowth, as well as metabolic responses to nutritional changesin the environment. As a consequence of their metabolisms,these four organisms differently contributed and competed fornutrients in the gut, which among those were short chain fattyacids, amino acids, and gasses.
DISCUSSION
Dysbiosis is one of the main features observed in metformin-treated T2D patients, where there is higher relative abundanceof Escherichia spp., A. muciniphila, S. variabile but lower ofI. bartlettii (Forslund et al., 2015; Mardinoglu et al., 2016; Wuet al., 2017). Moreover, larger concentrations of the SCFAspropionate and butyrate were reported under drug treatment(Forslund et al., 2015; Mardinoglu et al., 2016; Wu et al., 2017).However, the observation that metformin-treated T2D patientsshow a depletion in Firmicutes bacteria including I. bartlettii(Forslund et al., 2015; Wu et al., 2017), and that Firmicutes andClostridia are major sources of butyrate (Tremaroli and Bäckhed,2012; Shoaie et al., 2013, 2015), raises questions about the possiblesources of SCFAs. Systems biology approaches have consistentlybeen applied to clarify complex biological processes (Benfeitaset al., 2017; Lee et al., 2017, 2018; Uhlen et al., 2017) includingin the relationship between host and gut microbiota (Shoaieet al., 2013, 2015; Forslund et al., 2015; Mardinoglu et al., 2015).Here, we used systems biology methodologies including genome-scale metabolic models and flux balance optimization to clarifythe metabolic relationships between the prevalent gut bacteriaEscherichia sp., A. muciniphila, S. variabile, and I. bartlettiiand their contributions for extracellular pool of compoundsincluding SCFAs and amino acids. Based on synthetic lethality,we also examined the influence of uptake reactions, whichinvolve substrate exchange with the extracellular space, not onlyon bacterial growth rates but also on flux distribution acrossintracellular pathways.
The cumulative evidence presented here suggests thatthe shifts in microbiota diversity reported under metformintreatment and their resulting increase in butyrate and propionatepool (Forslund et al., 2015; Mardinoglu et al., 2016; Wuet al., 2017) may be due to an increased abundance ofS. variabile, a butyrate-producing anaerobe (Louis and Flint,2009). A. muciniphila may produce aminobutyrate, whileI. bartlettii produces isobutyrate, a branched chain fatty acidthat has been associated with increased risk of colon cancer(Shoaie et al., 2015). In turn, while the enzyme-coding genesinvolved in butyrate production are present in Escherichia sp.,this compound is not produced by the wild-type bacterium butmay be engineered to do so (Baek et al., 2013). It remains to testif other butyrate-producing bacteria (Forslund et al., 2015) showsimilar trends. Moreover, our modeling simulations indicate thatI. bartlettii, A. muciniphila, and Escherichia sp. may contributefor the extracellular pool of propionate, of which A. muciniphila
Frontiers in Physiology | www.frontiersin.org 10 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 11
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
FIGURE 5 | Contribution and competition of the four bacteria for extracellular substrates including short chain fatty acids (A), amino acids and nucleobases (B), andgases (C). Metabolic models for all organisms considered biomass maximization under minimal media anaerobic or aerobic conditions (respectively, solid anddashed arrows, see section “Materials and Methods” and Supplementary Tables S7, S8). Cys–Gly indicates Cysteinylglycine. Glutamine, deoxyribose, aspartate,and Cys–Gly show no flux under aerobic growth. Metabolite colors indicate the bacteria that influence its extracellular levels. For all bacteria we considered minimalmedium conditions (see section “Materials and Methods”) with exception to A. muciniphila, which additionally requires mucin to grow (de la Cuesta-Zuluaga et al.,2017).
had previously been observed to produce propionate (Derrienet al., 2004). These observations were associated with themajor changing pathways in the four organisms. Propionatemetabolism was the pathway displaying the highest responsesto alterations in nutrient uptake in I. bartlettii, together with
metabolism of phenylalanine. In turn, butyrate metabolism inS. variabile was perturbed depending on the inhibited reactions.
Bloating and intestinal discomfort are reported side-effectsof metformin medication (Forslund et al., 2015), and gassesproduced by gut microbiota enhance these adverse side effects
Frontiers in Physiology | www.frontiersin.org 11 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 12
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
(Lewis and Cochrane, 2007; Jahng et al., 2012; Forslundet al., 2015). Colonic transit may be beneficially influencedby the production of hydrogen (Lewis and Cochrane, 2007;Jahng et al., 2012). Our observations show that hydrogenwas produced by Escherichia sp., which also contributed forhydrogen sulfide production together with A. muciniphila.Hydrogen sulfide may have several beneficial effects forboth host and gut microbes, displaying anti-inflammatoryproperties, and promoting smooth muscle relaxation andantioxidant defense (Al Khodor et al., 2017). Future workshould experimentally test the contributions of the fourbacteria to the extracellular pool of gasses, SCFAs, and aminoacids.
The mechanism of action of metformin on glucosemetabolism was suggested to be mediated through thebacterial production of SCFAs, where local LPS-triggeredinflammation and lower intestinal lipid absorption are sideeffects of the drug (Forslund et al., 2015; Mardinoglu et al.,2016; Wu et al., 2017). A. muciniphila and Escherichia sp., twobacteria whose abundance is increased in metformin-treatedT2D patients, were the two organisms with the most sharedsingle and double essential reactions as indicated by syntheticlethality. These similar responses among the two bacteriaare consistent with the high sensitivity of LPS biosynthesis(a pathway exclusive of both bacteria), and nucleotide sugarmetabolism.
Among those nutrients that confer the highest increasesto Escherichia sp. growth both under anaerobic and aerobicgrowth are NMN and nitrate exchange, as well as tyrosine,phenylalanine, and tyramine uptake under aerobic conditions.NMN exchange is involved in the coenzyme nicotinamideadenine dinucleotide (NAD) salvage pathway I (Henry et al.,2010; Keseler et al., 2013; Wattam et al., 2017) essentialfor microbial catabolism and growth (Berríos-Rivera et al.,2002). In turn, nitrate exchange was the only reaction thatstimulated similar growth rate under alternative circumstances.Denitrification occurs as part of anaerobic respiration byreplacing oxygen as final electron acceptor in the electrontransport chain. Nitrate:nitrite antiporters (NarU and NarK)are responsible for the incorporation of nitrate and exportof nitrogen (Moreno-Vivian et al., 1999; Keseler et al.,2013). The catabolism of aromatic amino acids is one ofthe important commensal functions between this bacteria andthe host (Díaz et al., 2001; Fuchs et al., 2011), and plays animportant role in microbial-mediated food digestion in theintestine (Donaldson et al., 2015; Shoaie et al., 2015). Ourobservations further suggest that potential commensal behaviormay be displayed by Escherichia sp. and I. bartlettii underanaerobic growth, where on one hand the former producesphenylalanine required by the latter, and on the other handI. bartlettii produces proline, glutamine, and glutamate thatare uptaken by Escherichia sp. Moreover, phenylalanine,tryptophan, and threonine are essential amino acids whichmust be ingested by the host for nutritional availability,and which are part of the set minimal sufficient sourcespromoting cellular growth of Escherichia sp. Complementarily,arginine, cysteine, glutamine, glycine, proline, and tyrosine
are conditionally non-essential amino acids as well ascontributing as first line of sufficient sources for Escherichiasp. growth.
The dysbiosis induced by metformin treatment of T2Dpatients promotes nutritional imbalances (Forslund et al., 2015;Mardinoglu et al., 2016) that may impose fitness disadvantagesfor specific bacterial taxa. The alterations in relative abundanceof Escherichia sp. was consistent with our observed growthorganism requirements. Escherichia sp. is a facultative aerobe anddisplays a slightly lower number of essential uptake reactionsand higher number of uptake reactions when compared with theother bacteria under study. Additionally, the former organismis capable of growing while requiring fewer uptake reactionswhen compared with the other three organisms. Thus, ourobservations are consistent with Escherichia sp. showing ahigher robustness to environmental nutrient changes. Togetherwith its aerotolerance and steep oxygen gradient in the gut(Díaz et al., 2001; Bueno et al., 2012; Donaldson et al., 2015),this may confer a selective advantage to Escherichia sp. overother gut microbes (Díaz et al., 2001; Magnusdottir et al.,2017), allowing it to grow near the oxygen-rich epithelialsurface.
Interestingly, among all exchange reactions, simultaneousblocking of oxygen and ferric iron (Fe3+) uptake preventsgrowth of Escherichia sp, whereas ferrous iron (Fe2+) uptakeis by itself essential. Although iron and other metals may betoxic due to radical formation by reaction with reactive oxygenspecies [i.e., Fenton reaction (Koppenol, 1993)], it is essentialfor bacterial growth. Iron is a component of hemic enzymessuch as hydroperoxidases and cytochromes, and sensed by theBasS-BasR two-component system involved in LPS modificationand anoxic redox control (Bueno et al., 2012). In the absenceof oxygen, iron may act as electron acceptor whereby reductionof Fe3+ is coupled with oxidation of organic matter (Lovleyand Phillips, 1986), and its addition to cell culture promotesgrowth under anoxia (Bueno et al., 2012). Although one mayquestion whether the observed O2/Fe3+-associated lethalitypatterns are plausible considering that Fe2+ iron is uptakenby the cell, the oxidation of Fe2+ to Fe3+ by bacterioferritinrequires oxygen (4Fe2+
+ 4H+ + O2 → 4Fe3++ 2H2O).
The essentiality of iron in Escherichia sp. has been extensivelydiscussed elsewhere (Braun and Braun, 2002), and is encodedinto the biomass equation of Escherichia sp. where both redoxforms are present.
Overall, our in silico observations suggest commensaland competing behavior in the production of extracellularcompounds including short chain fatty acids and aminoacids, among which the metabolism of Escherichia sp.,A. muciniphila, S. variabile, and I. bartlettii may explainthe observed features in metformin-treated type 2 diabetespatients. These observations remain to be experimentallytested, though the above observations indicate good agreementwith previously known features of these organisms; multiplestudies have shown that growth predictions by FBA andgene essentiality prediction are in good agreement withexperimental observations (Edwards et al., 2001; Feist et al.,2007). Microbiota modulation approaches based on probiotics,
Frontiers in Physiology | www.frontiersin.org 12 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 13
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
prebiotics, and postbiotics are considered as potential therapiesin type 2 diabetes patients. Thus, identification of intestinalbacteria playing a beneficial role or promoting adverse effects onglucose and fatty acids metabolism, will allow the identificationof potential microbial targets to improve host metabolism.
AUTHOR CONTRIBUTIONS
AM conceived and supervised the study. DR, RB, GB, andCZ designed the experiments. DR performed the experiments.SS assisted in model acquisition and refining of the model.DR and RB analyzed the data and wrote the manuscript.All authors have revised and contributed to the finalmanuscript.
FUNDING
This work was funded by Knut and Alice Wallenberg Foundationand King’s College London.
SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be foundonline at: https://www.frontiersin.org/articles/10.3389/fphys.2018.00775/full#supplementary-material
TABLE S1 | Organism and AGORA reconstruction details.
TABLE S2 | Single and double essential reactions in Escherichia sp. andIntestinibacter bartlettii.
TABLE S3 | Flux changes in reactive oxygen species (ROS) reactions of I. bartlettiiunder aerobic and anaerobic conditions.
TABLE S4 | Influence of oxygen exchange reaction on Synthetic Lethality Analysisof I. bartlettii.
TABLE S5 | Organism exclusive and shared cellular metabolic pathways.
TABLE S6 | Flux variability analysis for the four bacteria.
TABLE S7 | Short chain fatty acids and metabolite consumption (negative flux)and production (positive flux) under maximal growth and minimal mediaconstraints.
TABLE S8 | Minimal media constraints applied on Escherichia sp. and I. bartlettiimodels.
REFERENCESAgren, R., Liu, L., Shoaie, S., Vongsangnak, W., Nookaew, I., and Nielsen, J. (2013).
The RAVEN toolbox and its use for generating a genome-scale metabolic modelfor Penicillium chrysogenum. PLoS Comput. Biol. 9:e1002980. doi: 10.1371/journal.pcbi.1002980
Al Khodor, S., Reichert, B., and Shatat, I. F. (2017). The microbiome and bloodpressure: can microbes regulate our blood pressure? Front. Pediatr. 5:138.doi: 10.3389/fped.2017.00138
Baek, J. M., Mazumdar, S., Lee, S. W., Jung, M. Y., Lim, J. H., Seo, S. W., et al. (2013).Butyrate production in engineered Escherichia coli with synthetic scaffolds.Biotechnol. Bioeng. 110, 2790–2794. doi: 10.1002/bit.24925
Benfeitas, R., Uhlen, M., Nielsen, J., and Mardinoglu, A. (2017). New challenges tostudy heterogeneity in cancer redox metabolism. Front. Cell Develop. Biol. 5:65.doi: 10.3389/fcell.2017.00065
Berríos-Rivera, S. J., Bennett, G. N., and San, K.-Y. (2002). The effect of increasingNADH availability on the redistribution of metabolic fluxes in Escherichiacoli chemostat cultures. Metab. Eng. 4, 230–237. doi: 10.1006/mben.2002.0228
Braun, V., and Braun, M. (2002). Iron transport and signaling in Escherichia coli.FEBS Lett. 529, 78–85.
Brillat-Savarin, A. (1826). You are what you eat. Physiol. Gout 32, 243–245.doi: 10.1002/cind.786-16.x
Bueno, E., Mesa, S., Bedmar, E. J., Richardson, D. J., and Delgado, M. J.(2012). Bacterial adaptation of respiration from oxic to microoxic and anoxicconditions: redox control. Antioxid. Redox Signal. 16, 819–852. doi: 10.1089/ars.2011.4051
de la Cuesta-Zuluaga, J., Mueller, N. T., Corrales-Agudelo, V., Velasquez-Mejia,E. P., Carmona, J. A., Abad, J. M., et al. (2017). Metformin is associated withhigher relative abundance of mucin-degrading Akkermansia muciniphila andseveral short-chain fatty acid-producing microbiota in the gut. Diabetes Care40, 54–62. doi: 10.2337/dc16-1324
Derrien, M., Vaughan, E. E., Plugge, C. M., and De Vos, W. M. (2004).Akkermansiamuciniphila gen. nov., sp. nov., a human intestinal mucin-degradingbacterium. Int. J. Syst. Evol. Microbiol. 54, 1469–1476. doi: 10.1099/ijs.0.02873-0
Díaz, E., Ferrández, A., Prieto, M. A., and García, J. L. (2001). Biodegradation ofaromatic compounds by Escherichia coli. Microbiol. Mol. Biol. Rev. 65, 523–569.doi: 10.1128/MMBR.65.4.523
Donaldson, G. P., Lee, S. M., and Mazmanian, S. K. (2015). Gut biogeography of thebacterial microbiota. Nat. Rev. Microbiol. 14, 20–32. doi: 10.1038/nrmicro3552
Edwards, J. S., Ibarra, R. U., and Palsson, B. O. (2001). In silico predictions ofEscherichia coli metabolic capabilities are consistent with experimental data.Nat. Biotechnol. 19, 125–130. doi: 10.1038/84379
Eker, S., Krummenacker, M., Shearer, A. G., Tiwari, A., Keseler, I. M., Talcott, C.,et al. (2013). Computing minimal nutrient sets from metabolic networks vialinear constraint solving. BMC Bioinformatics 14:114. doi: 10.1186/1471-2105-14-114
Feist, A. M., Henry, C. S., Reed, J. L., Krummenacker, M., Joyce, A. R., Karp, P. D.,et al. (2007). A genome-scale metabolic reconstruction for Escherichia coli K-12MG1655 that accounts for 1260 ORFs and thermodynamic information. Mol.Syst. Biol. 3:121. doi: 10.1038/msb4100155
Forslund, K., Hildebrand, F., Nielsen, T., Falony, G., Le Chatelier, E., Sunagawa, S.,et al. (2015). Disentangling type 2 diabetes and metformin treatment signaturesin the human gut microbiota. Nature 528, 262–266. doi: 10.1038/nature15766
Fuchs, G., Boll, M., and Heider, J. (2011). Microbial degradation of aromaticcompounds — from one strategy to four. Nat. Rev. Microbiol. 9, 803–816.doi: 10.1038/nrmicro2652
Henry, C. S., Dejongh, M., Best, A. A., Frybarger, P. M., Linsay, B., and Stevens, R. L.(2010). High-throughput generation, optimization and analysis of genome-scale metabolic models. Nat. Biotechnol. 28, 977–982. doi: 10.1038/nbt.1672
Hyotylainen, T., Jerby, L., Petaja, E. M., Mattila, I., Jantti, S., Auvinen, P., et al.(2016). Genome-scale study reveals reduced metabolic adaptability in patientswith non-alcoholic fatty liver disease. Nat. Commun. 7:8994. doi: 10.1038/ncomms9994
Imielinski, M., Belta, C., Rubin, H., and Halasz, A. (2006). Systematic analysisof conservation relations in Escherichia coli genome-scale metabolic networkreveals novel growth media. Biophys. J. 90, 2659–2672. doi: 10.1529/biophysj.105.069278
Jahng, J., Jung, I. S., Choi, E. J., Conklin, J. L., and Park, H. (2012). The effectsof methane and hydrogen gases produced by enteric bacteria on ileal motilityand colonic transit time. Neurogastroenterol. Motil. 24, 185–191. doi: 10.1111/j.1365-2982.2011.01819.x
Ji, B., and Nielsen, J. (2015). From next-generation sequencing to systematicmodeling of the gut microbiome. Front. Genet. 6:219. doi: 10.3389/fgene.2015.00219
Kaelin, W. G. Jr. (2005). The concept of synthetic lethality in the context ofanticancer therapy. Nat. Rev. Cancer 5, 689–698. doi: 10.1038/nrc1691
Karlsson, F. H., Tremaroli, V., Nookaew, I., Bergstrom, G., Behre, C. J.,Fagerberg, B., et al. (2013). Gut metagenome in European women with normal,impaired and diabetic glucose control. Nature 498, 99–103. doi: 10.1038/nature12198
Frontiers in Physiology | www.frontiersin.org 13 June 2018 | Volume 9 | Article 775
fphys-09-00775 June 21, 2018 Time: 15:56 # 14
Rosario et al. Modeling Gut Microbiota Nutritional Requirements
Keseler, I. M., Mackie, A., Peralta-Gil, M., Santos-Zavaleta, A., Gama-Castro, S.,Bonavides-Martínez, C., et al. (2013). EcoCyc: fusing model organism databaseswith systems biology. Nucleic Acids Res. 41, 605–612. doi: 10.1093/nar/gks1027
Koppenol, W. H. (1993). The centennial of the Fenton reaction. Free Radic. Biol.Med. 15, 645–651.
Lee, S., Zhang, C., Arif, M., Liu, Z., Benfeitas, R., Bidkhori, G., et al. (2018). TCSBN:a database of tissue and cancer specific biological networks. Nucleic Acids Res.46, D595–D600. doi: 10.1093/nar/gkx994
Lee, S., Zhang, C., Liu, Z., Klevstig, M., Mukhopadhyay, B., Bergentall, M., et al.(2017). Network analyses identify liver-specific targets for treating liver diseases.Mol. Syst. Biol. 13:938. doi: 10.15252/msb.20177703
Lewis, S., and Cochrane, S. (2007). Alteration of sulfate and hydrogen metabolismin the human colon by changing intestinal transit rate.Am. J. Gastroenterol. 102,624–633. doi: 10.1111/j.1572-0241.2006.01020.x
Louis, P., and Flint, H. J. (2009). Diversity, metabolism and microbial ecology ofbutyrate-producing bacteria from the human large intestine. FEMS Microbiol.Lett. 294, 1–8. doi: 10.1111/j.1574-6968.2009.01514.x
Lovley, D. R., and Phillips, E. J. (1986). Organic matter mineralization withreduction of ferric iron in anaerobic sediments. Appl. Environ. Microbiol. 51,683–689.
Magnusdottir, S., Heinken, A., Kutt, L., Ravcheev, D. A., Bauer, E., Noronha, A.,et al. (2017). Generation of genome-scale metabolic reconstructions for 773members of the human gut microbiota. Nat. Biotechnol. 35, 81–89. doi: 10.1038/nbt.3703
Mardinoglu, A., Boren, J., and Smith, U. (2016). Confounding effects of metforminon the human gut microbiome in type 2 diabetes. Cell Metab. 23, 10–12.doi: 10.1016/j.cmet.2015.12.012
Mardinoglu, A., Boren, J., Smith, U., Uhlen, M., and Nielsen, J. (2018). Systemsbiology in hepatology: approaches and applications. Nat. Rev. Gastroenterol.Hepatol. 15, 365–377. doi: 10.1038/s41575-018-0007-8
Mardinoglu, A., and Nielsen, J. (2012). Systems medicine and metabolic modelling.J. Intern. Med. 271, 142–154. doi: 10.1111/j.1365-2796.2011.02493.x
Mardinoglu, A., Shoaie, S., Bergentall, M., Ghaffari, P., Zhang, C., Larsson, E.,et al. (2015). The gut microbiota modulates host amino acid and glutathionemetabolism in mice. Mol. Syst. Biol. 11:834. doi: 10.15252/msb.20156487
Moreno-Vivian, C., Cabello, P., Martinez-Luque, M., Blasco, R., and Castillo, F.(1999). Prokaryotic nitrate reduction: molecular properties and functionaldistinction among bacterial nitrate reductases. J. Bacteriol. 181, 6573–6584.
O’Neil, N. J., Bailey, M. L., and Hieter, P. (2017). Synthetic lethality and cancer.Nat. Rev. Genet. 18, 613–623. doi: 10.1038/nrg.2017.47
Pratapa, A., Balachandran, S., and Raman, K. (2015). Fast-SL: an efficient algorithmto identify synthetic lethal sets in metabolic networks. Bioinformatics 31,3299–3305. doi: 10.1093/bioinformatics/btv352
Qin, J., Li, Y., Cai, Z., Li, S., Zhu, J., Zhang, F., et al. (2012). A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature 490, 55–60.doi: 10.1038/nature11450
Schellenberger, J., Thiele, I., and Orth, J. D. (2012). Quantitative prediction ofcellular metabolism with constraint-based models. Nat. Prot. 6, 1290–1307.doi: 10.1038/nprot.2011.308
Shoaie, S., Ghaffari, P., Kovatcheva-Datchary, P., Mardinoglu, A., Sen, P., Pujos-Guillot, E., et al. (2015). Quantifying diet-induced metabolic changes of thehuman gut microbiome. Cell Metab. 22, 320–331. doi: 10.1016/j.cmet.2015.07.001
Shoaie, S., Karlsson, F., Mardinoglu, A., Nookaew, I., Bordel, S., and Nielsen, J.(2013). Understanding the interactions between bacteria in the human gutthrough metabolic modeling. Sci. Rep. 3:2532. doi: 10.1038/srep02532
Tremaroli, V., and Bäckhed, F. (2012). Functional interactions between thegut microbiota and host metabolism. Nature 489, 242–249. doi: 10.1038/nature11552
Turanli, B., Grøötli, M., Borén, J., Nielsen, J., Uhlen, M., Arga, K. Y., et al. (2018).Drug repositioning for effective prostate cancer treatment. Front. Physiol. 9:500.doi: 10.3389/fphys.2018.00500
Uhlen, M., Zhang, C., Lee, S., Sjostedt, E., Fagerberg, L., Bidkhori, G., et al. (2017).A pathology atlas of the human cancer transcriptome. Science 357:eaan2507.doi: 10.1126/science.aan2507
Varemo, L., Scheele, C., Broholm, C., Mardinoglu, A., Kampf, C., Asplund, A.,et al. (2015). Proteome- and transcriptome-driven reconstruction of the humanmyocyte metabolic network and its use for identification of markers fordiabetes. Cell Rep. 11, 921–933. doi: 10.1016/j.celrep.2015.04.010
Wattam, A. R., Davis, J. J., Assaf, R., Boisvert, S., Brettin, T., Bun, C., et al.(2017). Improvements to PATRIC, the all-bacterial bioinformatics database andanalysis resource center. Nucleic Acids Res. 45, D535–D542. doi: 10.1093/nar/gkw1017
Wu, H., Esteve, E., Tremaroli, V., Khan, M. T., Caesar, R., Mannerås-Holm, L., et al.(2017). Metformin alters the gut microbiome of individuals with treatment-naive type 2 diabetes, contributing to the therapeutic effects of the drug. Nat.Med. 23, 850–858. doi: 10.1038/nm.4345
Zhang, C., and Hua, Q. (2016). Applications of genome-scale metabolic models inbiotechnology and systems medicine. Front. Physiol. 6:413. doi: 10.3389/fphys.2015.00413
Conflict of Interest Statement: The authors declare that the research wasconducted in the absence of any commercial or financial relationships that couldbe construed as a potential conflict of interest.
Copyright © 2018 Rosario, Benfeitas, Bidkhori, Zhang, Uhlen, Shoaie andMardinoglu. This is an open-access article distributed under the terms of the CreativeCommons Attribution License (CC BY). The use, distribution or reproduction inother forums is permitted, provided the original author(s) and the copyright ownerare credited and that the original publication in this journal is cited, in accordancewith accepted academic practice. No use, distribution or reproduction is permittedwhich does not comply with these terms.
Frontiers in Physiology | www.frontiersin.org 14 June 2018 | Volume 9 | Article 775
